Sensor data streams are prevalent in various real-time applications within the Internet of Things (IoT). However, these data streams often suffer from missing values due to issues like sensor malfunctions, communication failures, or drained batteries. Such missing data can negatively impact the quality of real-time analytics and downstream applications. Existing imputation methods often rely on strong assumptions about the data streams or lack efficiency. In this study, we aim to accurately and efficiently impute missing values in data streams that exhibit only general characteristics, thereby enhancing the effectiveness of real-time applications. By improving the MPIN model [11] - one of the state-of-the-art models for imputing missing IoT values, we introduce the MIMP model (Missing data Imputation by Message Propagation) with higher accuracy in performing imputation. This study will present how to tune the MPIN model and then provide experimental evaluations to demonstrate its effectiveness.